Decoder. plain-English AI glossary

Bias

● Core

Also called AI bias

When a model's outputs systematically favor or disfavor certain groups — usually because its training data did.

Think of it like

A mirror that doesn't reflect everyone equally — it shows some people clearly and distorts others.

Example

A hiring model trained mostly on male resumes starts ranking women lower — not because it was told to, but because the data leaned that way.

How it actually works

Bias enters at every stage: data collection (who's represented), labeling (whose judgments), model design (what's optimized), and deployment (who's affected). It's not just a technical bug — it reflects and can amplify real-world inequities.

For product teams

Systematic unfairness in AI output — a legal, ethical, and reputational risk that needs active monitoring.

For engineers

Systematic skew from data/labeling/optimization; mitigate via diverse data, audits, disaggregated eval.

Related

Read anything AI without the jargon

Look up any term in plain English, or save terms as you read with the free Chrome extension.

Open DecoderAdd to Chrome